gpu.latency

Submits a trivial probe through a swizzled Metal command queue on a fixed interval, timing how long it waits before the GPU starts it and how long it takes once started.


Delivery	stream, interval
Mechanism	swizzle
Requires	iOS 17+, hardware only, no simulator, basset 0.1.1+

Real output from basset request get, captured while drawing through a CAMetalLayer:

CAPTURED                  ELAPSED  INSTRUMENT

launch 5600321339416417470
2026-08-23T16:10:27.314Z  +0.0s    metal.gpu.latency(gpu)
├─ windowNanoseconds            999978250
├─ sampleCount                  1
├─ occurrenceCount              1
├─ frameBudgetNanoseconds       16666666
├─ gpuWaitNanoseconds           115042
├─ gpuWaitPeakNanoseconds       115042
├─ gpuExecutionNanoseconds      5583
└─ gpuExecutionPeakNanoseconds  5583
2026-08-23T16:10:28.313Z  +1.0s    metal.gpu.latency(gpu)
├─ windowNanoseconds            999274125
├─ sampleCount                  1
├─ occurrenceCount              1
├─ frameBudgetNanoseconds       16666666
├─ gpuWaitNanoseconds           278375
├─ gpuWaitPeakNanoseconds       278375
├─ gpuExecutionNanoseconds      5250
└─ gpuExecutionPeakNanoseconds  5250
2026-08-23T16:10:29.313Z  +2.0s    metal.gpu.latency(gpu)
├─ windowNanoseconds            999866541
├─ sampleCount                  1
├─ occurrenceCount              1
├─ frameBudgetNanoseconds       16666666
├─ gpuWaitNanoseconds           250750
├─ gpuWaitPeakNanoseconds       250750
├─ gpuExecutionNanoseconds      5542
└─ gpuExecutionPeakNanoseconds  5542

Getting Started

SDK

CLI

Instruments

Reference